Direct Policy Transfer via Hidden Parameter Markov Decision Processes